Jiang Jie, Vice President of Tencent: AI is improving efficiency in every aspect of advertising. Tencent will recruit more AI talents.

Text by Wang Yuchan

In the third quarter of this year, Tencent's advertising revenue reached a new high in the past six quarters with a growth rate of 21%. Advertisers in all major industries increased their advertising investments.

This achievement is not only due to the increase in ad loading rate but also the growth of eCPM driven by AI - powered ad targeting. The earnings report specifically mentioned the "Tencent Advertising AIM+" intelligent advertising - placement product matrix. It enables advertisers to automatically configure targeting, bidding, and ad positions, and optimize ad creatives, thereby enhancing their return on marketing investment.

Tencent stated that through AIM+, for every 10,000 yuan spent on advertising by advertisers, the number of operations required on the advertising platform decreased by 80%, and the number of operations in the creative process decreased by 47%.

As a strong competitor in the AI "arms race", Tencent has made substantial capital expenditures on AI. In 2024, Tencent's total capital expenditure increased by 221% year - on - year, and this expenditure will be even higher in 2025. Currently, one manifestation of the "return on investment" from AI is in Tencent's advertising business.

The high cost of AI is not only due to the need to reserve a large amount of computing power resources but also reflected in the competition for talent. The "2025 AI Talent Mobility Report" released by Maimai shows that from January to July 2025, the number of newly posted AI jobs increased by more than 10 times year - on - year, and the number of resume submissions also soared by 11 times. There is a continuous shortage of "algorithm - related" talents, with the shortage of "search algorithm" talents being the most severe, with five positions competing for two candidates. The number of non - technical positions increased by 7.7 times year - on - year.

In Maimai's statistics of the "Ranking of the Number of Newly Posted AI Jobs by Enterprises from January to July", Tencent ranked fifth. The top four were ByteDance, Xiaohongshu, Alibaba, and Ant Group.

In this competition, several companies have launched AI algorithm competitions, offering prizes and direct job offers to recruit talents in the market. On November 27, during the finals of the "Tencent Advertising Algorithm Competition", Jiang Jie (Zeus), the vice - president of Tencent, was interviewed by several media outlets including 36Kr after the competition. He discussed the boosting effect of AI on the advertising business, the future direction of technological evolution, and the consideration of AI talents from the perspective of the advertising business.

This "Tencent Advertising Algorithm Competition" offered a prize pool of 3.6 million yuan, direct job offers at Tencent, and practical opportunities based on real - world business scenarios, attracting more than 8,400 applicants from nearly 30 countries around the world. Approximately 75% of the final 20 finalists were post - 2000s, while most of the post - 1990s were doctoral students.

The following is the transcript of the dialogue. 36Kr has made deletions and adjustments to the content without changing the original meaning:

AI Improves Efficiency in Every Aspect of Advertising

Q: Under the current ad loading rate, how does AI improve the recommendation efficiency for advertisers and optimize the user experience?

Jiang Jie: The ad loading rate has little to do with the advertising system itself. The low loading rate reflects our restraint in commercialization. Currently, the ad loading rate of Tencent's Video Account is around 4%, while the industry average is generally between 10% and 15%, a three - fold difference. This is because Tencent has always been cautious about commercialization, which is also related to the WeChat ecosystem. We aim to provide users with a better experience.

Q: What changes might generative AI bring to ad recommendation? How will Tencent Advertising address the challenges and risks brought by these changes? How do you think the roles of advertising optimizers and traffic operators will evolve in the future?

Jiang Jie: In the future, there will be two approaches for the advertising system. One is the discriminative approach, and the other is the new generative approach.

The generative approach can solve the problem of ad cold - start. By leveraging large language models and world knowledge, it doesn't suffer from the lack of world knowledge like the traditional discriminative approach, which requires a large number of samples to solve the cold - start problem.

However, the generative approach cannot solve all problems. The traditional discriminative approach uses a multi - stage cascading architecture in retrieval/recall, rough ranking, and fine ranking (Note: The current discriminative advertising system is like a funnel, filtering candidate ads through the above - mentioned stages layer by layer and finally recommending one ad to the user). The generative approach is gradually catching up, and everyone is still exploring. Since the traditional discriminative advertising system has been developed for at least a decade or two, we have accumulated a lot of experience.

We are gradually replacing the recall and rough - ranking stages with the generative approach, and this approach has achieved significant results, which are also reflected in the revenue data in the earnings report. AI is improving efficiency in every aspect of advertising, including recommendation, creativity, and placement.

As for the future of advertising optimizers, I can clearly tell you that in the future, they will not be responsible for ad placement but for creating ad materials. This is a major transformation. Currently, we use intelligent placement to replace the tasks of price adjustment and bidding by optimizers. You don't need to set bids anymore because the system can find the optimal price and help you obtain the maximum volume, as the system has a global perspective. Optimizers will no longer need to monitor the market in the future, as Tencent Advertising AIM+'s intelligent placement can solve these problems.

In the future, optimizers will mainly focus on solving creativity problems. Many current AI tools for generating video and images can provide optimizers with a lot of capabilities to improve creative efficiency. However, I believe that Tencent Advertising will protect original creativity. I hope that optimizers will create more original works because originality cannot be achieved by AI alone but requires human creativity.

I believe that advertising optimizers will always exist, but their focus will change. In the future, it will be difficult to distinguish between ads and native videos. You can also obtain information and knowledge from ads. Ads and content will be inseparable in the future, which highlights the importance of originality.

Q: From the release of Gemini, it seems that there is no major obstacle to directly generating ad materials based on customer information. Is it just a matter of cost?

Jiang Jie: I think prompt optimization is still very important. It is still difficult to write a good prompt, which requires creativity and product relevance. The screening of prompt materials is also crucial, and these factors are strongly related to advertising optimizers.

The issue of latency is gradually being resolved. It used to take five minutes to generate a video, but now it can be done in five or ten seconds. The cost is still a problem, but it is gradually being addressed. Previously, the generated videos were large, but small models will have more application scenarios in the future. For example, Hunyuan has launched an 8B model, and there is also a 1B OCR model that was just tested yesterday, which is better than DeepSeek. It is small but powerful, with low inference costs and a wide range of future application scenarios. This is a trend in the future development of AI.

Q: Tencent's earnings report for the first quarter of this year mentioned that "with the help of AI optimization, the click - through rate of some ad inventory has increased to around 3.0%." (Historical data: The click - through rate of banner ads is about 0.1%, and that of in - feed ads is about 1.0%). Is there any new progress or data regarding AI optimization of the click - through rate?

Jiang Jie: Currently, the ability to achieve "one - to - one creative for each user" cannot generate real - time creative content, but it can generate images. We are also developing interactive video capabilities. The goal of "one - to - one creative for each user" can be achieved in the future. Once the inference cost and efficiency of machines improve, the cost of generative content will become low enough.

In fact, our current advertising system already supports multiple creatives for one ad. In the future, as the number of creatives increases, there will be more options. It is entirely possible for one ad to have ten or even hundreds of creatives in the future, provided that the cost of multi - modal generation decreases.

Tencent Advertising Miaosi can already generate a large number of ad creatives through AIGC. The production efficiency of short - video digital human materials has been improved: The daily output of materials by merchants in the industry has increased from 20 pieces per person per day to 60 pieces per person per day, with a 300% increase in labor efficiency. The comprehensive production cost of materials has been reduced by 50%, and the material production efficiency has been increased by 50%. In terms of effectiveness, the customer penetration rate has reached 65%. 3 million Miaosi creatives are distributed and applied every day, and approximately 47% of the effective creatives for WeChat Mini - Program ads come from Miaosi.

I believe that in the future, when displaying an ad, we can generate a video in real - time based on the user's immediate commercial interests. While the user is browsing, we can generate interactive ad materials in real - time, enabling strong interaction with the content. The traditional cascading approach will be replaced by an end - to - end approach, which depends on the continuous reduction of inference costs.

Q: For Tencent, AI has indeed improved the effectiveness and efficiency of advertising. Other brands are also using AI. Does the use of AI expand the overall advertising market, or does it intensify competition?

Jiang Jie: Advertisers' annual advertising budgets remain relatively stable. AI will only accelerate technological innovation and improve the user experience. Users will stay where the experience is better. AI mainly improves efficiency, but the total annual marketing budget generally remains unchanged; only the form has changed. Undoubtedly, it will lead to greater competition. Competition drives progress, and as everyone explores, it will accelerate the overall technological transformation. I think this is a good thing.

Q: What cutting - edge technologies or business projects is Tencent currently exploring and will implement in the future?

Jiang Jie: I think cutting - edge technologies mainly include the following aspects: Starting with large models, first, the ability of large language models (LLM), represented by DeepSeek and OpenAI's GPT - related models, is to improve the upper limit of intelligence. Gemini can solve the most difficult competition and doctoral - level problems, and the upper limit of AI intelligence is constantly being broken, similar to when AlphaGo played Go, and no human could defeat it later.

In the future, the upper limit of AI will be gradually broken. Previously, we conducted pre - training, which was a simulated training. For example, if you show a camera to AI, it only knows it is a camera but doesn't know it is a camera on a tripod. However, with reinforcement learning, its intelligence upper limit increases, and it can generalize and distinguish between a three - legged and a four - legged camera. This is an improvement in the upper limit of intelligence. When the RL reinforcement training reaches 10 steps, 100 steps, or 200 steps, the upper limit of intelligence is different. We have made significant improvements in mathematical and coding abilities, which is a technological change.

Next is multi - modality. In the future, multi - modality will be the key support. All native multi - modal editing capabilities for images have been completed, and one model can solve all problems. We have also open - sourced Image3.0. In the future, VEO3 and Sora2 will solve the native multi - modal capabilities for videos. The 3D world model is very valuable for many automobile companies. Previously, automobile companies had to collect a large amount of real - world data with a wide range of collection angles. AI can help collect this data more precisely. The collection angle has changed from once every 15 degrees to once every degree.

The world model is also applied in the virtual world of games, and technological changes are rapid. Tencent has been continuously iterating on basic capabilities. We are working on all these aspects, including the 3D world model, which will also be applied in the advertising system.

The AI Talent Market is a "Seller's Market", and Top - Tier Talent Needs to Be Competed for

Q: Is there anything different about this year's competition compared to previous years? Have you seen anything unexpected or surprising from the young participants this year, whether in their solutions or personal qualities?

Jiang Jie: The algorithm competition has been held since 2017, almost for a decade. I was deeply impressed after listening to the solutions presented by the young participants in this year's competition. We will hire more young people.

I was pleasantly surprised to find that the understanding and knowledge of large models among current students in school are very close to what is being done in the industrial sector. Unlike when we were students, we had to learn from a tutor for half a year, gradually understand the programming environment, and then become familiar with the process. In fact, current students don't need to go through such a process at all. This is a fundamental difference. Their knowledge system is completely in line with ours (in the industrial sector) and is even more innovative, which really surprised me.

Yesterday, I told several participants that they only used the desensitized data in the competition, with millions of data points in the preliminary round and tens of millions in the semi - final round. If they join Tencent and work on real - world business, I believe they will achieve better results. This is why I hope they can quickly join our team.

Q: What unique value can Tencent offer to the participants who receive job offers from Tencent?

Jiang Jie: I think there are several aspects. First, Tencent has a diverse range of businesses and a rich ecosystem, which is probably the most comprehensive among current Internet companies. When young people work on large models at Tencent, they can engage in more fields, whether it's multi - modality or advertising.

Second, for outstanding students in the advertising algorithm competition, we will definitely offer them the "Qingyun Program", which is the top - tier recruitment channel for students, ensuring them good compensation and future incentives. Tencent has a dedicated Qingyun training system with high - end mentors and resources.

This is what we can offer to students after they join Tencent: a high - quality practical environment, good treatment, and a comprehensive training system.

After the algorithm competition, I will communicate more with these students. Just as we did with the Qingyun students, we will invite them to visit Tencent, communicate and dine with them, and let them have in - depth communication with Tencent's culture and future department leaders. This is also very important. Our interaction with them should not be limited to the competition. More interaction is needed to retain them at Tencent, rather than having them leave after receiving the prize money. This is not a normal form of communication, and neither we nor they can learn anything from it. More interaction is valuable.

Q: Since last year, we have seen fierce competition among large companies and emerging AI startups in recruiting AI talent. Some companies offer stock options and high salaries. Even Alibaba and ByteDance have their own "Qingyun Programs". How do you view this round of competition for AI talent? What are the fundamental differences compared to the previous era, such as the Internet era? What qualities do we value most in candidates?

Jiang Jie: The previous era was the mobile Internet era, which we experienced. We all grew up with the mobile Internet era. I deeply remember that about 20 years ago, during the mobile Internet era, when companies recruited young people, they only cared if the candidates graduated from a good university and had relevant majors. They didn't care whether the candidates had interned at a large company.

I entered the Internet industry by pure luck. At that time, most people joined IBM or other large foreign companies. Companies mainly recruited people with good engineering skills, such as those who could use C++ and JAVA.

Now, I value candidates' abilities more. The candidates recruited through the "Qingyun Program" have at least interned at large companies and have worked on many projects at startups. When we were graduates, we had to undergo a one - month training in the new - employee training class to understand the company's culture. However, these new candidates are already familiar with our culture and can start working immediately.

They have a rich knowledge system. Some have studied statistics, some are in IT, and some are from mathematics departments. Their skills are highly interactive, as they understand both algorithms and engineering. The competition in large language models ultimately comes down to engineering capabilities, but the skill requirements for engineering are different. This is a significant difference.

Previously, when we launched a large project, it required the cooperation of many people. Now, one person can handle it, or at most, a small team is needed. This was difficult in the past. Students had to follow senior experts and integrate a lot of resources to complete a project. Now, the situation is completely different.

Now, the infrastructure is different, and candidates' abilities are also different. Currently, a team of one to three people can do what a large team used to do. This is a fundamental difference.